The Talairach daemon is in Talairach space, of course, and not in MNI space. If you do want to use it in MNI space, then you might want to consider using a method to transform the dataset to Talairach space, as in our previous thread using 3dfractionize.
If you do want to use this atlas in its native space, you would change the 3dresample command to have a -master of the dataset that you want to mask, instead of using the -dxyz option.
The Talairach daemon atlas, however, has several issues. First, it has no template available, so the TT_N27 dataset does not align particularly well with it. Consequently, all transformations (3dfractionize uses a specific warp between an MNI N27 and TT N27) provide only a very approximate transformation. The atlas was done for a specific person with no available image data. Secondly, the atlas is very coarse with 5mm spacing between slices. Thirdly, the digitized version of the atlas both introduces and exposes various errors - cross slice discontinuities and outright digitizing mistakes. Finally, the Brodmann regions, in particular, are drawn very finely on cortical boundaries that typically lie outside the brain even in Talairached datasets, making alignment overlap difficult So while we provide this atlas because of its historical importance in AFNI and to the general neuroscience community, it is not recommended.
One of the main advantages of the Talairach atlas is its Brodmann regions, that are only partially described in other atlases. I do have an alpha version of a Brodmann atlas stored in the link below. This was developed by Juergen Mai and Milan Matajnik. I believe this provides a much better mapping of the Brodmann regions. These are in the MNI 2009c asymmetric space, projected from the MNI 2009c cortical surfaces into the cortical ribbon mask.
